Analysis · Digital economy

The real business of artificial intelligence

A critical reading of generative AI's still unresolved economic model.

The dispute between the United States government and Anthropic, the company behind Claude, opened the door to a set of questions that usually remain outside the public conversation. The money flowing into artificial intelligence is astronomical, but the business model for the final consumer is still unresolved.

The contradiction is visible everywhere. Generative AI products have millions of users, dominate technology headlines and receive enormous investments from the most powerful companies in the world. Yet it remains unclear who will pay for them at the scale required to sustain the infrastructure behind them.

Artificial intelligence is expensive before it becomes profitable. Training frontier models requires chips, energy, specialized engineers, data centers and continuous experimentation. Running those models for everyday users also has a cost, because every prompt consumes computing power.

That makes AI different from many digital businesses that scaled through near-zero marginal costs. A social network can serve one more post or one more image very cheaply. A large language model, by contrast, spends real resources each time a user asks it to think, summarize, code or reason.

The consumer subscription model is one answer, but not necessarily the definitive one. Many people try the tools, some pay for premium access, and companies bundle AI into existing products. Still, the gap between public enthusiasm and sustainable revenue is a central tension in the industry.

Another answer is enterprise adoption. Companies can use AI to automate tasks, improve customer service, write software, analyze documents and accelerate internal workflows. That market may be larger and more stable than individual subscriptions, but it requires integration, trust and measurable productivity gains.

The military and national security uses of AI reveal another layer of the business. Governments are interested in systems that can process intelligence, simulate scenarios, support logistics and improve decision-making. In that world, the customer is not the casual user but the state itself.

This is why the ethical debate around Anthropic matters. The company built part of its identity around AI safety, constitutional principles and responsible development. But the frontier between civilian tools and government or defense uses is increasingly difficult to draw.

For the companies involved, the incentives are enormous. Big Tech firms are not investing in AI only because it is fashionable. They are trying to own the next layer of computing: the models, the cloud infrastructure, the chips, the operating systems and the interfaces through which people will work.

Microsoft, Google, Amazon, Meta, Nvidia and OpenAI are all part of a wider reorganization of power. Some control cloud infrastructure. Some control chips. Some control distribution. Some control the models. The real business may be the full stack rather than a single chatbot subscription.

That is why strategic alliances have become so important. A model company needs cloud capacity. A cloud company needs a model that makes its infrastructure indispensable. A software company needs AI inside its productivity tools. Each agreement reshapes the competitive map.

There is also a data problem. The first wave of models was trained on vast amounts of public and licensed material. Future improvement may depend on higher-quality data, synthetic data, private enterprise data and specialized domains. Access to those sources will become a business advantage.

The public debate often focuses on whether AI will replace jobs. That question is important, but it can obscure another one: who captures the economic value produced by AI? Workers may become more productive, but platforms may capture the margin if they own the tools and distribution.

In that sense, AI could repeat an old pattern from the digital economy. Technology promises decentralization and empowerment, but the infrastructure becomes concentrated in a small group of companies with capital, talent and market access.

The current moment is not simply a race to build smarter machines. It is a race to define the terms under which intelligence as a service will be sold, governed and embedded into institutions. The winners will not only sell software; they will shape how work is organized.

Regulation enters the scene late but necessarily. Governments want innovation, productivity and strategic advantage, but they also worry about monopolies, misinformation, labor disruption, security and dependency on private platforms.

The challenge is that regulation moves at the speed of law while AI moves at the speed of capital. By the time a rule is written, a new model, a new business agreement or a new technical capability may have changed the terrain.

For consumers, the question remains practical. Will AI be a paid assistant, an invisible layer inside existing software, a public utility, an enterprise service, an advertising business, or a combination of all of those? The answer is still being negotiated.

What is already clear is that artificial intelligence is not only a technological revolution. It is an economic project. The companies building it are not merely creating tools; they are building markets, dependencies and political influence.

The real business of AI may therefore be less about charging for prompts and more about controlling the infrastructure of the next digital era. The chatbot is the visible surface. The deeper business is the operating system of work, knowledge and decision-making.

Originally published in Ámbito.